help
help > RE: regression as anova
Jul 16, 2015 05:07 PM | Alfonso Nieto-Castanon - Boston University
RE: regression as anova
Hi Liron,
My apologies for the late reply (we have busy preparing next week's conn workshop).
I may be misinterpreting what you mean by an "ancova contrast", the [0 0 0 1 0 0 0 0] is in my mind exactly that, a contrast the looks at the 'emotion' covariate effect in an ancova model that includes both a categorical 'group' factor with three levels, and a 'emotion' continuous covariate (as well as other covariates of no interest). Perhaps what you are after is instead simply a contrast like:
[1 -1 0 0 0 0 0 0;
0 1 -1 0 0 0 0 0;
0 0 0 1 0 0 0 0]
That will look at the ROIs that show either a between-group difference or an association with the emotion covariate in a single test (instead of performing two separate tests, one to identify the group effects and another to identify the emotion effects)?
Please let me know if that is what you had in mind (and otherwise please give me more details about what information you would like to get from your desired ancova contrast to see if that helps me better understand what you are after).
Best
Alfonso
Originally posted by L R:
My apologies for the late reply (we have busy preparing next week's conn workshop).
I may be misinterpreting what you mean by an "ancova contrast", the [0 0 0 1 0 0 0 0] is in my mind exactly that, a contrast the looks at the 'emotion' covariate effect in an ancova model that includes both a categorical 'group' factor with three levels, and a 'emotion' continuous covariate (as well as other covariates of no interest). Perhaps what you are after is instead simply a contrast like:
[1 -1 0 0 0 0 0 0;
0 1 -1 0 0 0 0 0;
0 0 0 1 0 0 0 0]
That will look at the ROIs that show either a between-group difference or an association with the emotion covariate in a single test (instead of performing two separate tests, one to identify the group effects and another to identify the emotion effects)?
Please let me know if that is what you had in mind (and otherwise please give me more details about what information you would like to get from your desired ancova contrast to see if that helps me better understand what you are after).
Best
Alfonso
Originally posted by L R:
Thank you so much for this fast and elaborated
answer!
The contrasts you specified are the contrasts I actually used in the first place. I used the results as a screening step (to identify the ROI-pairs that are significantly correlated with group or emotion), and then entered group+emotion+covariates into a single ancova model in spss using exported data from CONN (sorry for not explaining that in the original massage, I thought it was long and complicated as is), to compare the relative contribution of group and emotion (each after controlling for all the rest, i.e. group after controlling for covariates+emotion, and vice verca).
The issue is that in the "screening step", i.e. looking at CONN results, I use two different methods to identify the roi-pairs. I was advised to use the same method for both, even though the regression results should be the same as ancova results (i.e. regression of emotion and connectivity controlling for group and covariates should be the same as an ancova that takes into account emotion+group+covariates when looking the the contribution of emotion in predicting connectivity). Theoretically, I could have just entered all into a single ancova model in CONN and looked at the contribution of group and emotion separately (like I do in spss), but CONN gives a single value- the value for a specific covariate after controlling for all all others.
For this reason I would like, if possible, to find an ancova contrast that is equivalent to the regression contrast that you specified:
"b) contrast [0 0 0 1 0 0 0 0] to test for Emotion effects while controlling for all other effects (Group & covariates)"
Thank you very much for your help!
Liron.
The contrasts you specified are the contrasts I actually used in the first place. I used the results as a screening step (to identify the ROI-pairs that are significantly correlated with group or emotion), and then entered group+emotion+covariates into a single ancova model in spss using exported data from CONN (sorry for not explaining that in the original massage, I thought it was long and complicated as is), to compare the relative contribution of group and emotion (each after controlling for all the rest, i.e. group after controlling for covariates+emotion, and vice verca).
The issue is that in the "screening step", i.e. looking at CONN results, I use two different methods to identify the roi-pairs. I was advised to use the same method for both, even though the regression results should be the same as ancova results (i.e. regression of emotion and connectivity controlling for group and covariates should be the same as an ancova that takes into account emotion+group+covariates when looking the the contribution of emotion in predicting connectivity). Theoretically, I could have just entered all into a single ancova model in CONN and looked at the contribution of group and emotion separately (like I do in spss), but CONN gives a single value- the value for a specific covariate after controlling for all all others.
For this reason I would like, if possible, to find an ancova contrast that is equivalent to the regression contrast that you specified:
"b) contrast [0 0 0 1 0 0 0 0] to test for Emotion effects while controlling for all other effects (Group & covariates)"
Thank you very much for your help!
Liron.
Threaded View
| Title | Author | Date |
|---|---|---|
| L R | Jul 8, 2015 | |
| Alfonso Nieto-Castanon | Jul 9, 2015 | |
| L R | Jul 10, 2015 | |
| Alfonso Nieto-Castanon | Jul 16, 2015 | |
| L R | Sep 30, 2015 | |
| Alfonso Nieto-Castanon | Sep 30, 2015 | |
| Alfonso Nieto-Castanon | Sep 30, 2015 | |
| L R | Jul 21, 2015 | |
| Alfonso Nieto-Castanon | Jul 24, 2015 | |
| L R | Jul 24, 2015 | |